N EXPERIMENTAL TEST ON DYNAMIC CONSUMPTION AND LUMP SUM …
Transcript of N EXPERIMENTAL TEST ON DYNAMIC CONSUMPTION AND LUMP SUM …
AN EXPERIMENTAL TEST ON DYNAMIC CONSUMPTION AND LUMP‐SUM
PENSIONS1
Authors
ENRIQUE FATÁS JUAN A. LACOMBA FRANCISCO LAGOS ANA I. MORO‐EGIDO LINEEX
Campus Tarongers Universidad de Valencia
46022 Valencia Spain
Departament of EconomicsFaculty of Economics University of Granada Campus La Cartuja s/n
18071 Granada Spain
Phone +34963828643 Fax +34963828415 [email protected]
Phone +34958249605Fax +34958249995 [email protected] [email protected] [email protected]
ABSTRACT
This article examines the potential risks on consumption behavior of lump‐sum payments in public pension systems. Lump‐sum payments can be consumed too fast and generate an increase of poverty rates. We experimentally investigate retirement consumption behavior in an inter‐temporal decision‐making setting. Subjects make consumption and saving decisions in an environment with two central features: first, there exists a decreasing probability of survival; and second, in addition to the regular income they get while active, they receive a unique lump‐sum payment when retired. The results of this paper indicate that in the vast majority of periods subjects over‐save. Moving the lump‐sum payment into earlier periods makes the effect eventually deeper: subjects adopt larger precautionary saving decisions. JEL category: C91, H55, J26. Keywords: Experimental test, consumption, savings, lump‐sum payment.
1 We thank people from CREED (Amsterdam) for helpful comments during the elaboration of the experimental design and seminar participants at Malaga, Tucson, Granada and Rome. Financial support from the Spanish Ministry of Education and Science, the Regional Government of Andalucia and Fundación Centro de Estudios Andaluces is gratefully acknowledged.
I. INTRODUCTION
The reform of Social Security systems is now one of the main issues on the
economic policy agenda of most industrialized countries. It is widely considered that,
unless serious changes take place, the aging of the population implying a rise in the
number of retirees relative to that of workers will threaten the viability of Pay‐As‐You‐Go
public pension systems in the long‐run.
Besides, this threat is being reinforced by the progressive reduction in the
retirement age of the working population. Pension systems in virtually all OECD countries
in the mid‐1990s made it financially unattractive to work after the age of 55.1 Indeed, the
general consensus in the theoretical literature related to Social Security and retirement
decisions is that pension systems create enormous incentives to leave the labour force
early. 2 This large decline in labour force participation is attributed to the specific fact that
to keep on working implies a reduction in the present value of total pension benefits. That
is, it is considered that the drop in pension wealth acts as an implicit tax on income from
continued work and as such is a clear incentive to retire early.
Reforms aiming to increase the effective retirement age to improve the financial
problems of public pensions systems have mainly focused on the reduction of this implicit
tax on prolonging the working period.
It is considered that when the increase in pension benefits is exactly offset by the
higher cost in terms of contributions and foregone pensions, the pension system is not
distorting the retirement decision. That is, the pension systems that are marginally
actuarially fair do not distort the individual retirement decisions. For this reason, the main
economic policy measures move in the direction of strengthening the link between life‐
time contributions and pension benefits.3
However, Cremer, Lozachmeur and Pestieau (2006) argue that “while there is no
doubt that retirement systems induce an excessive bias towards early in many countries, a
complete elimination of this bias (i.e., a switch to an actuarially fair system) is not the right
answer. This is so and for two reasons. First, some distortions are second‐best optimal.
Second, depending on the political process, it may either not be feasible or alternatively it
may tend to undermine the political support for the pension system itself.” 4
Therefore, a key question is whether or not there exist alternative reforms to
increase the effective retirement age. Orszag (2001), related to U.S. Social Security,
considered that transforming Social Security's delayed retirement credit (given to people
working between the ages of 62 and 65 in the U.S.) into a lump‐sum payment rather than
an increased monthly payment would likely encourage people to defer retirement.
This question is addressed in Fatas, Lacomba and Lagos (2007). They find that the
more concentrated the payments (shifting from annuity into lump‐sum), the more
postponed the retirement decisions. This results suggests, in the line of Orszag (2001), that
reforms aimed to delay effective retirement ages should transform the increases in
pensions due to the additional years of work (after the standard retirement age) into a
lump‐sum payment rather than an increased periodic payment. Additionally,
Fetherstonhaugh and Ross (1999), using a questionnaire, find evidence that people would
be more willing to delay retirement if they received a lump‐sump payment rather than an
increased annuity.
Furthermore, it is likely that this transformation including a lump‐sum payment
would be easier to implement. In the U.S. private industry, whose retirement benefits may
be distributed in several alternative ways, using some type of lump‐sum benefit as a
payment option has become popular as an alternative to annuity payments (see Moore
and Muller, 2002, Blostin, 2003 or Butrica and Mermin, 2006).
However, the incorporation of a lump‐sum payment as a measure to delay
retirement decisions requires further analysis before receiving full consideration by
policymakers. As Orszag (2001) also states, paying lump‐sum payments might result in
increases in poverty rates among those who already delay their retirement decisions after
the normal retirement age if the lump sums were mostly consumed rather than saved.
Little is known about how quickly retirees would spend down their lump‐sum
payments, or how these reforms would affect consumption patterns. Butrica and Mermin
(2006), using data from the Health and Retirement Study (HRS), examine how household
expenditures among adults aged 65 and older vary by the degree of annuitization. They
find that if Social Security was completely privatized, and retirees did not annuitize,
discretionary spending could increase by as much as 22 percent for married adults and 38
percent for unmarried adults. However, other studies have shown that consumption
usually declines at retirement. This decline has increasingly been referred to as the
retirement consumption puzzle. See Fisher et al (2005) for references.
These opposite results suggest that more research is needed before unambiguous
conclusions can be made. This is precisely the aim of this work: to provide additional
empirical evidence to this debate. To this end, we consider an experimental investigation
to test the potential effects on consumption behavior of implementing a lump‐sum
payment in a public pension system. We also analyze how closely the predictions of the
optimality theory fit the actual behavior of subjects in a lab.
To our knowledge, this is the first experimental approach to examine a dynamic
saving‐consumption problem in a retirement framework. The experimental design is based
on two central features: first, there exists a decreasing probability of surviving which
implies an uncertain future income; and, secondly, there are two sequences of income,
one when individual works and another when she is retired.
The results of this paper indicate that subjects over‐react to the lump‐sum
payments, as they briefly consume above their optimal dynamic consumption paths.
However, and on aggregate, the general picture goes in the opposite direction. In the vast
majority of periods they over‐save rather than over‐consume. This result holds for both the
periods prior to the lump‐sum payment and no income periods, when smoothing becomes
more difficult. Moving the lump‐sum payment into earlier periods (so, increasing the
number of periods with no income at all) seems to have a counterintuitive effect. Subjects
tend to over‐react even more and adopt precautionary saving measures. Our results
suggest that Social Security reforms aimed at moving from traditional annuitized pensions
to lump‐sum payments might not yield increases in poverty rates.
This paper is organized as follows. Section 2 briefly surveys the experimental
literature on this topic. Section 3 develops the theoretical model, while section 4 presents
the experimental design and procedures. Section 6 presents our results and the last one
concludes.
II. EXPERIMENTAL BACKGROUND
In the experimental literature into consumption behaviour under uncertainty there
exist few contributions but relevant ones. For instance, Hey and Dardadoni (1988) describe
a large‐scale experimental investigation to test the implications of expected utility
maximization on optimal consumption behaviour. They find that the actual behaviour in
the lab differs significantly from the optimal behaviour, and that the comparative static
implications of actual behaviour agree with those of optimality theory. Carbone and Hey
(2004) investigate the over‐sensitivity of consumption to current income. They adopt a
simple model in which income in any period can take just one of two values: employment
and unemployment. Unlike neoclassical theoretical predictions about smooth consumption
over time, they find that subjects over‐react. That is, their results show a close relationship
between consumption and current income.
On the other hand, Ballinger et al (2003) study social learning about the life cycle
saving task. They use experimental methods to study a household inter‐temporal choice
problem. Subjects participate in three‐member “families”. Second and third “generation”
subjects observe and/or communicate with their “antecedent” first or second generation
subject. They find that later generations perform significantly better than earlier
generations. Brown, Camerer and Chua (2006) establish potential ways in which
consumers can attain near‐optimal consumption behaviour. In line with Ballinger et al
(2003), individual and social learning mechanisms are proposed to be one possible link.
They find that while consumers persistently spend too much in early periods, they learn
rapidly from their own experience and from experience of others to consume amounts
close to optimal levels.
Our work presents some interesting differences with regard to all above mentioned
papers. In order to capture a retirement and pension system benchmark, we focus on how
subjects make saving and consumption choices with two novel features. First, subjects face
a decreasing probability of surviving across periods. Second, participants receive three
different levels of income according to a retirement period (R hereafter): i) a constant level
of income during each period before R (as worker); ii) a higher lump‐sum of income in R
(the first period as retired); and iii) nothing from R on.
III. THE MODEL
Before we proceed with the experimental design and results, in this section we are
going to characterize the optimal consumption decisions using a standard model. Consider
an individual who has to decide on his optimum consumption at different age in presence
of uncertainty about the length of life. Suppose that this is the only uncertainty that the
individual faces. Let T>0 be the planning horizon, that is, maximum lifetime. Lifetime
uncertainty is presented by a survival distribution function, F(t) non increasing in age, 0 ≤ t
≤ T, that satisfies F(0)=1.5 All individuals are assumed to have the same survival distribution
function.
Denote consumption at age t by c(t), where c(t)≥0. Utility from consumption at
different ages is separable and independent of age. There is no subjective discount rate.6
We use a specific utility function, u(c) that displays risk‐aversion, and to ensure an interior
solution, satisfies the Inada conditions.7 When working, the individual provides one unit of
labor. Contingent on survival, the individual works between ages 0 and R, 0<R<T, that is,
there exists a mandatory retirement age that occurs at R. The individual's objective
function is to maximize the non discounted expected utility V
)).(u( F(t)T
1t
tcV ∑=
= (1)
Let wages at age t, be w(t)≥0. Savings earn a zero rate of interest.8 With no initial
assets, the individual's assets at age t, S(t), are equal to the cumulative savings. Feasible
consumption plans must have non‐negative assets at all ages
∑∑==
=t
1j
R)min(t,
1j)( -w(j))( jctS . (2)
We assume that w(t)=w for all 0<t<R, therefore the restriction becomes
[ ] ∑=
−=t
1j
)( -,)1(min)( jctwwRtS . (3)
The choice of the optimum consumption path will depend on the insurance options
available. Relative to this benchmark case, we analyze an alternative scenario, in which
individuals will receive a unique lump‐sum payment at the beginning of the retirement
period. The total value of expected contributions has to be equal than the lump‐sum
payments, that is,
LS )( w F(t)1-R
1t
RF=∑=
τ , (4)
then the lump‐sum payment is
)( w)(
F(t) w
1-R
1t RHRF
LS ττ ==∑= . (5)
Restriction on savings is now
[ ] ],1[)( -)1(,min)(t
1j
RtjcwtLStS ∈∀−= ∑=
τ . (6)
This theoretical model will be used to measure to which extent subjects deviate
form the optimal consumption path in the experiment.
IV. EXPERIMENTAL DESIGN AND PROCEDURES
Our experimental design tries to capture some actual features of an actuarially fair
public pension system with a unique lump‐sum payment.9 The experiment consists of
three sequences of at most 30 rounds and one decision per round. See Table 1 into
Appendix.
‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐ Insert Table 1 here ‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐
Each round is characterized by a probability of surviving. As the round number
increases, the probability of surviving decreases. In each round reached, subject either
survives or not. A subject reaches a round if, and only if, she has survived all earlier rounds.
During the first R‐1 rounds, similar to wage earnings, subject receives 85
experimental units in each reached round (Income).
In round R, subjects receive the present value of the total pension benefits as a
unique lump‐sum payment. As a strategy to analyze the effect of the length of the
retirement period on savings and spending decisions, we design two treatments. In
Treatment 1 (hereafter LS10) the round R is the 10th and in Treatment 2 (hereafter LS15)
the round R is the 15th.
We consider a gross wage of 100 experimental units and a tax rate of 15%. Thus, in
order to reflect an actuarially fair pension system, the lump‐sum payment in LS10 is equal
to 191.25 experimental units (with R=10) and in LS15 is equal to 345 experimental units
(with R=15) following equation (5).
In the remaining rounds subjects receive no income at all. Therefore, in LS10 and in
LS15 subjects have at most 19 and 14 rounds with no income, respectively.
In each round subjects have to make a unique decision about how to divide the
Available Cash into consumption and savings. The available cash comes from the addition
of the income received in that period and what is not consumed in previous rounds (their
cumulative savings). So, income can be saved to provide wealth but savings earn no
interest and borrowing is not allowed, that is, subjects cannot spend more than their
Available Cash.
Let C denote the amount of experimental units converted into points by subjects in
each round (their consumption). Subjects are informed of the conversion scale (from
experimental units to points, converted into real euros at the end of the experiment): C
experimental units generate 20*Square Root (C) points. A table mapping how different
consumption choices are converted into points (euros) is given separately to subjects.
As mentioned above, subjects play three sequences. They are told that at the end
of the experiment they will be privately paid in cash the total amount of points converted
from experimental units of one of the three sequences (randomly chosen). They are also
told that any unconverted experimental unit remaining at the end of any sequence is
worthless.
A total of 39 undergraduate students in Business and Economics from the
University of Valencia took part in the experiment. All sessions were run at the Laboratory
for Research in Experimental Economics (LINEEX) and standard electronic recruitment
procedures were used to collect the subject pool. All subjects participating in the
experiment had previously participated (weeks before the experiment was run) in other
sessions facing a Raven test, a risk aversion test and a basic socioeconomic survey.10 The
list of participants in every session was randomized to control for any reputation or group
identity effect. Anonymity was preserved using random codes and these tests were paid
independently.
At the beginning of the experiment, subjects entered the laboratory and were
randomly seated in a private cubicle. Experimental instructions were read aloud by the
experimenter (instructions are provided in the Appendix). To make sure subjects
understood the logic of the game, subjects completed a quiz before the experiment began.
Explanations were repeated until all subjects passed the quiz (nobody made a mistake
almost from the beginning, quiz available from authors upon request).
At the end of the experiment subjects were privately paid with an exchange rate of
125 experimental units = €1 for one of the three sequences, chosen at random. On average
an experimental session lasted less than 90 minutes, the average earnings were around
€27 and the maximum earnings peaked above €40.
The experiment consisted of two treatments: LS10 and LS15. 20 subjects
participated at LS10 and 19 subjects at LS15.
V. RESULTS
Table 2 presents the summary statistics of the individuals’ decisions about
consumption, their cumulative savings and available cash at the beginning of each period,
for both the LS10 and LS15 treatments, by sequences. The number of survived periods in
the last row denotes the average number of rounds that subjects were alive and making
decisions. Average life is almost identical in both treatments (21.96 versus 21.31, as
expected in a pure randomly driven process).
By pure inspection, average consumption is higher in LS15 than in LS10. This follows
the experimental design, as subjects get a higher lump‐sum and get a positive income
during a higher number of periods in LS15. However, the more interesting information
regarding treatment effects comes from the next two variables. Both the average
cumulative savings and the available cash are a direct consequence of subjects’ decisions
in both treatments. Again, by pure inspection, subjects save considerably more in LS10
(152.8) than in LS15 (101.4): the difference is around 50%. This suggests that subjects in
the worst scenario are able to save more to protect themselves from the higher number of
periods with no income. Moreover, the average available cash is larger in LS10, in spite of
the fact that available cash is the sum of cumulative savings and income (by definition,
larger in LS15 than in LS10). Over‐saving behaviour in LS10 more than compensates this
difference.
No clear differences are observed across sequences within each treatment in
consumption and savings. The actual consumption and saving choices do not show a clear
(increasing or decreasing) trend as participants play more sequences. However, the
standard deviation falls as subjects make more decisions. Besides, this reduction in the
standard deviation across sequences seems to be systematic for all available variables.
That is, in a very preliminary way, subjects seem to adjust better as the number of
decisions increases.
The descriptive information shown above is not enough to answer our main
research question. In order to learn about subjects’ capabilities to get close to the optimal
consumption path, we need to measure the distance between actual decisions and our
theoretical benchmark.
We include two measures of divergence between actual and optimal behaviour. In
our view, there are two plausible ways of thinking about optimality in this context. The first
way is to define optimal consumption as the level of consumption calculated in the first
period for the rest of the life‐time profile. We will call this optimal consumption the ex‐
ante optimal consumption (EAOC). To compute EAOC we solve the dynamic optimization
problem in period zero to find the optimal smoothed consumption path. The policy
consequences of this approach to optimal decisions are straightforward: EAOC would be
the rational choices done by a benevolent social planner.
The second approach to optimality takes as the natural analysis unit the per‐round
individual decision. For every subject and every available cash amount, based on her
previous decisions, we compute the optimal consumption for every period, contingent on
the fact that (i) she is alive and (ii) owns a given actual positive wealth. This alternative
approach is a kind of natural basis for assessing the optimality of subjects’ decisions
dynamically. We call this ex‐post optimal consumption (EPOC).
Note that EAOC and EPOC have different meanings. The natural interpretation for
EAOC comes from the policy side: it is a general average measure of the adjustment of the
population (in our case, the subject pool) to the optimal path in welfare terms. Significant
deviations in this ex ante measure can be associated to welfare losses. EPOC gives
additional valuable information about the optimality of decisions at the individual level.
For every subject, in every period, it defines the optimal decision, contingent on the fact
that she is still making decisions over her individual level of wealth.
To facilitate the comparison of actual and optimal consumption levels over all
individuals we plot them. Figure 1 and 2 (3 and 4) correspond to average consumption and
cumulative savings in LS10 (LS15). In Figures 1 and 3, the actual consumption paths are
compared with EAOC and EPOC, while the associated optimal cumulative savings are used
in Figures 3 and 4.
In all cases both a risk aversion rate of .50 and a risk aversion rate of .28 are used to
check whether our results are robust to changes in risk aversion rates. The reasons for this
duality are twofold: on the one hand, a constant risk aversion rate of .50 is implicitly used
in the payoff function converting points into earnings.11 In addition, it was the mode
obtained in our risk aversion test. On the other hand, a risk aversion coefficient of .28 is
the average constant risk aversion rate of our subjects, measured by our risk aversion test.
[Figures 1 and 3 around here]
Unlike previous experimental papers, we find no evidence of over consumption in
the first rounds. On the contrary, our participants under‐consume in this earliest periods.12
This under‐consumption behaviour lasts until round 5 in LS10 and round 10 in LS15 using
the EAOC as a benchmark, while it holds in almost all rounds when the EPAC measure is
used. The results are qualitatively identical for both risk aversion measures.
From an ex ante perspective, and before the lump‐sum period comes, consumption
exhibits some inertia and subjects fail to smooth it. Relative to the EAOC path, subjects
over consume temporarily around the lump‐sum period, in rounds 8 to 11 in LS10 and
rounds 11 to 15 in LS15. However, this over consumption disappears when EPOC is used as
the reference point. With the exclusion of the very same period when the lump sum is
received, and only under a high risk aversion rate, subjects under consume in all periods
relative to the optimal path.
Interestingly, after the lump‐sum payment period, when subjects get no income at
all, participants in the experiment performed relatively well adjusting their behaviour to
the optimal path. Rather to consume their available income immediately after the lump‐
sum period, they keep on smoothing their consumption and adjust it to the optimal path in
a more than reasonable way.
A natural way to back up these descriptive comments is to have a look at the
evolution of cumulative savings over time in Figures 2 and 4. Subjects perform saving
choices in a relatively poor manner, in the sense that they over‐save in all periods in LS10,
regardless of the risk aversion rate used in the analysis. In LS15 subjects also over‐save in a
vast majority of periods, with the notable exception of periods around the lump‐sum
payment, when their reaction to the sharp increase in income is not compensated by a
parallel jump in savings. This general behavioural pattern does not change after receiving
the lump‐sum payment. It suggests that savings tend to be precautionary, with the
exception of the periods in which the high lump‐sum is received.
[Figures 2 and 4 around here]
Before we proceed with a more formal analysis of the experimental data it is worth
to note that our design can be considered as a kind of a strong test for over‐saving (and
under‐consumption). On the one hand, in Hey and Dardadoni (1988) and Carbone and Hey
(2004) the rate of return per period is known and certain, for all money saved. On the
other, in Ballinger et al (2003) and Brown, Camerer and Chua (2006) the incentives to save
money are embedded in their particular utility function, so saving is salient for subjects.
In our experimental design savings yields no returns. Our utility function, though
concave, differs of the ones used by Ballinger et al (2003) and Brown, Camerer and Chua
(2006) to avoid the salience of saving money. Additionally, subjects were not punished in
any specific way when their consumption was critically low, as no negative utility was
associated to a zero consumption level. We additionally paid very much attention to keep
framing as neutral as possible to avoid any kind of demand effect.
In a more formal way, we run some regressions to explore the determinants of
decisions. The dependent variable is the log of absolute deviation from optimality (both
relative to EAOC and EPOC: LD‐EAOC and LD‐EPOC). A negative coefficient means that the
corresponding independent variable lowers the deviation from optimality. Since the
dependent variable is the logged deviation, the coefficient means that a variable causes a
certain percentage of increase or decrease in deviation relative to when this variable is
absent.
We consider different groups of explanatory variables. Dumbreak is a dummy that
takes the value of 1 for all LS10 data, 0 otherwise, and it intends to capture the existence
of any treatment effect. Seq2 (Seq 3) takes the value of 1 when data is generated in the
second (third) sequence of 30 rounds.
As the probability of survival decreases with the number of periods played, we
include both a Period and a Period Squared variable. Hand in hand with the increasing
uncertainty (measured by the decreasing probability of survival), we expect participants to
make worse decisions as the number of periods increases, so we expect a positive
coefficient for this variable. The Period Squared variable simply takes into account any
possible non‐linearity that the period variable may have.
Risk is the risk aversion of subjects, as measured by the Holt and Laury risk aversion
test. The original value range (from 0 to 10), is decomposed into three values: 0 if subject is
risk loving, 1 if she is risk neutral and 2 if she is risk averse. Raven Test is the ratio of correct
answers in the test (a non verbal IQ test) run by all subjects participating in the
experiment. Sex is a dummy variable to control for gender effects (0 for females, 1 for
males).13 We report the estimation results in Table 3.
[Table 3 around here]
The first result emerging from Table 3 is that timing matters when dealing with
lump sum schemes. Postponing the retirement period and getting the lump sum later,
significantly increases the deviation from the optimal path. Even when subjects get a
positive income fewer times in LS10, they are relatively better able to cope with the
optimization problem and smooth their consumption over time.
No sequence dummy is significant, so our results suggest that there is very little
learning across stages, if any. Coefficients on Period and Period Squared show that, as
expected, deviations increase as more periods are played, but at a decreasing rate. Risk
aversion is significant. Risk‐averse subjects outperform risk lovers (as the negative
coefficients show). Surprisingly enough, a better performance in the non‐verbal
intelligence test (the “Raven test”) is related to an increase in the distance to optimal
consumption levels. We do not have a solid explanation for this finding. Finally, males
appear to deviate more in almost all cases.
Following Brown, Camerer and Chua (2006), we want to check whether a natural
behavioural explanation for the observed patterns of consumption is the existence of a
“rule of thumb”. Subjects simply spend a fixed fraction of their current income or a fraction
of available cash. To investigate this alternative explanation, we ran regressions in which
the log of actual consumption is regressed against the optimal level of consumption and
either current income or available cash (i.e, income plus cumulative savings). Table 4
summarizes these results.14
[Table 4 around here]
Table 4 suggests that subjects mainly focus on their current income when dealing
with consumption decisions. Besides, the coefficient of Income is larger than that of
Available cash, which means that subjects decide their consumption levels regardless of
their savings. This is consistent with the over‐saving behavioural pattern. The increment in
the R‐squared value is small when considering Available cash but large in the case of
Income.
VI. CONCLUSIONS
One of the proposals to delay the effective retirement age is to transform the
increases in pensions due to the additional years of work into a lump‐sum payment rather
than an increased periodic payment. This idea was first suggested by Orszag (2001) for the
U.S. Social Security. He considered that transforming Social Security's delayed retirement
credit (given to people working between the ages of 62 and 65 in the U.S.) into a lump‐sum
payment would likely encourage people to defer retirement.
However, this measure risks increasing poverty rates. People who do not annuitize
much of their retirement wealth might spend too quickly the lump‐sum payment reducing
income at very old ages. In this paper we have tested in the laboratory the potential
effects on consumption behavior of implementing a lump‐sum payment. Our main
conclusion is that this might not be the case. The introduction of a unique lump‐sum
payment generates a behavioral overreaction in the opposite direction: rather than
consuming too much, too fast, our subjects show a persistent precautionary saving
behavior.
The policy consequences of our experiment are that transforming (at least,
partially) pension benefits into lump‐sum payments might not increase elderly poverty
rates.
An additional risk, pointed by Orzsag (2001), comes from the political pressure to
extend lump‐sum payments to those who are younger than the regular retirement age.
Such an extension would raise the risk of increasing elderly poverty rates. The analyses of
results in our LS10 treatment, relative to the LS15, suggest the opposite: the earlier and
the lower the lump‐sum payment, the stronger the saving reaction.
This paper is the first step on the experimental analysis of consumption behavior
during retirement. Although we believe that our findings might help to achieve proper
Social Security reforms, we are also aware of its limitations. For instance, as Orszag (2001)
states, one important issue is whether the lump‐sum payment should be an option
available in addition to the existing scheme, or should simply replace it. If the system is
voluntary, selection effects across alternatives could raise costs to the Social Security
system (i.e., those with shorter life expectancies could disproportionately choose the
lump‐sum payment, while those with longer life expectancies could disproportionately
choose the annuity option). A choice between the two systems should be incorporated in
future experimental treatments of the problem. This is the topic of our ongoing research.
VII. REFERENCES
1. Ballinger T. P., Palumbo G. M. and T.N. Wilcox. Precautionary Saving and Social
Learning across Generations: an experiment. Economic Journal, 113, 2004, 920‐947.
2. Blondal, S. and S. Scarpetta. “The retirement decision in OECD countries.” OECD
Economic Department Working Papers, Nº202, OECD Publishing, 1998.
3. Blostin, A.P. “Distribution of retirement income benefits.” Monthly Labor Review, 126
(4), 2003, 3‐9.
4. Breyer, F. and M. Kifmann. “Incentives to retire later: a solution to the Social Security
crisis?” Journal of Pension Economics and Finance, 1, 2002, 111‐130.
5. Brown, A., Camerer, C. and Z. Chua. “Learning and visceral temptation in dynamic
savings experiments” . Working Paper, 2006.
6. Butrica, Barbara and Gordon B.T. Mermin. ”Annuitized wealth and consumption at
older ages”, The Center for Retirement Research WP #2006‐26.
7. Carbone, E. and J. Hey. "The effect of unemployment on consumption: an experimental
analysis." Economic Journal, Royal Economic Society, 114 (497), 2004, 660‐683.
8. Casamatta, G., H. Cremer and P. Pestieau. “Voting on Pensions with Endogenous
Retirement Age.” International Tax and Public Finance, 12, 2005, 7‐28.
9. Camerer, C. and Z. Chua. Experiments on Intertemporal Consumption with Habit
Formation and Social Learning. Working Paper, 2005.
10. Conde‐Ruiz J.I. and V. Galasso. “Early Retirement.” Review of Economics Dynamics, 6,
2003, 12‐36.
_____ “The Macroeconomics of Early Retirement.” Journal of Public Economics, 88,
2004, 1849‐1869.
11. Conde‐Ruiz J.I., V. Galasso and P. Profeta, 2003. “Economic Development, Aging and
Retirement Decisions.” Mimeo, 2003.
_____ “The Evolution of Retirement.” CEPR Discussion Papers, n. 4683, 2005.
12. Coller, M. and M.B. Williams. “Eliciting Individual Discount Rates.” Experimental
Economics, 2, 1999, 107‐127.
13. Crawford, V.P. and D.M. Lilien. “Social Security and the Retirement Decision.” Quarterly
Journal of Economics, 96, 1981, 505‐529.
14. Cremer, H., J. Lozachmeur and P. Pestieau. “Social Security and Retirement Decision: A
Positive and Normative Approach.” CEPR Discussion Paper no. 5542, 2006, London,
Centre for Economic Policy Research.
15. Cremer, H. and P. Pestieau. “Reforming Our Pension System: Is It a Demographic,
Financial or Political Problem?” European Economic Review, 44, 2000, 974‐983.
_____ “The Double Dividend of Postponing Retirement.” International‐Tax‐and‐Public‐
Finance, 10 (4), 2003, 419‐34.
16. Crettez, B. and P. Le Maitre. “Optimal Age of Retirement and Population Growth.”
Journal of Population Economics, 15, 2002, 737‐755.
17. Cubitt, R.P. and R. Sudgen. “Dynamic Decision‐Making Under Uncertainty: An
Experimental Investigation of Choices Between Accumulator Gambles.” The Journal of
Risk and Uncertainty, 22:2, 2001, 103‐128.
18. Fatas, E., J.A. Lacomba and F. Lagos. “An experimental test on retirement decisions.”
Economic Inquiry, 45 (3), 2007, 602‐614.
19. Fehr, H., W.I. Sterkeby and O. Thorgersen. “Social security reforms and early
retirement.” Journal of Population Economics, 16, 2003, 345‐361.
20. Feldstein, M.S. “Social security and private savings: international evidence in an
extended life‐cycle model,” in The Economics of Public Services, Proceedings of a
Conference held by the International Economics Association, edited by M.S. Feldstein
and R.P. Inman, 1977.
21. Fetherstonhaugh, D. and Lee Ross. “Framing Effects and Income Flow Preferences in
Decisions About Social Security,” in Behavioral Dimension of Retirement Economics,
edited by Aaron. Washington D.C, 1999.
22. Fisher, Jonathan, David S. Johnson, Joseph Marchand, Timothy M. Smeeding, and
Barbara Boyle Torrey. ”The Retirement Consumption Conundrum: Evidence from a
Consumption Survey”, The Center for Retirement Research WP #2005‐14.
23. Frederik, S., G. Loewenstein and T. O’Donoghue. “Time Discounting and Time
Preferences: A Critical Review.” Journal of Economic Literature. 40(2), 2002, 351‐401.
24. Gruber, J. and D. Wise, eds. “Social Security Programs and Retirement around the
world,” in University of Chicago Press, 1997.
25. Hamermesh, D. and P.L. Menchik. “Planned and Unplanned Bequests.” Economic
Inquiry, 25(1), 1987, 55‐66.
26. Harrison, G.W., L. Morten and M.B. Williams. “Estimating Individual Discount Rates for
Denmark: A Field Experiment.” American Economic Review, 92(5), 2002, 1606‐1617.
27. Herbertsson, T.T. and M. Orszag. “The Cost of Early Retirement in the OECD.” IoES
Working Paper, No. W01:02, 2001.
28. Hey, J. And V. Dardadoni. “Optimal Consumption under Uncertainty: An Experimental
Investigation.” Economic Journal, 98 (390), 1987, 105‐16
29. Holt, C. and S.K. Laury. “Risk Aversion and Incentive Effects.” American Economic
Review, 92(5), 2002, 1644‐1655.
30. Kotlikoff, L.J. “Social Security and Equilibrium Capital Intensity.” Quarterly Journal of
Economics, 93, 1979a, 233‐53.
_____ “Testing the theory of social security and life cycle accumulation.” American
Economic Review, 69, 1979b, 396‐410.
31. Lacomba, J.A. and Francisco Lagos. “Population Aging and Legal Retirement Age.”
Journal of Population Economics, 19 (3), 2006, 507‐519.
_____ “Political Election on Legal Retirement Age.” Social Choice and Welfare, 29 (1),
2007, 1‐17.
32. Laibson, D.I. “Golden Eggs and Hyperbolic Discounting.” Quarterly Journal of Economics
112, 1997, 443‐447.
33. Moore, J.H. and L.A. Muller. “An analysis of lump‐sum pension distribution recipients.”
Monthly Labor Review, 125 (5), 2002, 29‐46.
34. O’Donoghue, T. and M: Rabin. “Doing it Now or Doing it Later.” American Economic
Review, 89, 1999, 103‐124.
35. Organization For Economic Cooperation and OECD (2000). Maintaining Prosperity in an
Ageing Society. Organisation for Economic Co‐operation and Development.
36. Orzag, P. “Should a lump‐sum payment replace social security’s delayed retirement
credit?” An issue in brief. Center for Retirement Research at Boston College, 2001,
Number 6.
37. Samwick, Andrew A. “Discount Rate Heterogeneity and Social Security Reform.” NBER
Working Paper No. W6219, 1997.
38. Sheshinski, E. “A Model of Social Security and Retirement Decisions.” Journal of Public
Economics, 10, 1978, 337‐360.
39. Social Security Administration. “The Impact of Repealing the Retirement Earning Test
on Rates of Poverty”. Office of Policy, 2000, February 23.
40. Steuerle, E. and J. Bakija. “Retooling Social Security for the 21sstt Century: Right and
Wrong Approaches to Reform.” Washington, DC: Urban Institute Press, 1994.
41. Thaler, Richard. “Intertemporal Choice.” In Richard Thaler, ed., The Winner’s Curse:
Paradoxes and Anomalies of Economic Life. Princeton, NJ: Princeton University Press,
1992.
VIII. APPENDIX
TABLE 1:EXPERIMENTAL DESIGN
Rounds Pobability of surviving
LS10 Income
LS15 Income
Available Cash
Consumption Savings Obtained Points
1 1 85 85 85 ‐‐‐ ‐‐‐ ‐‐‐
2 29/30 85 85
3 28/29 85 85
4 27/28 85 85
5 26/27 85 85
6 25/26 85 85
7 24/25 85 85
8 23/24 85 85
9 22/23 85 85
10 21/22 85 85
11 20/21 191,25 85
12 19/20 85
13 18/19 85
14 17/18 85
15 16/17 85
16 15/16 345
17 14/15
18 13/14
19 12/13
20 11/12
21 10/11
22 9/10
23 8/9
24 7/8
25 6/7
26 5/6
27 4/5
28 3/4
29 2/3
30 1/2
ºº 0,00
TABLE 2: DESCRIPTIVE STATISTICS LS10 LS15
Averages Total Seq = 1 Seq = 2 Seq = 3 Total Seq = 1 Seq = 2 Seq = 3Consumption 43,82 49,32 40,27 43,74 66,22 65,89 68,11 65,11 [44,20] [50,51] [44,61] [36,62] [44,65] [50,35] [44,18] [40,47] Cum. Savings 152,8 148,0 149,6 161,4 101,4 111,8 89,4 102,5 [187,35] [198,69] [187,81] [176,36] [135,31] [144,99] [128,05] [132,75] Available cash 196,6 197,3 189,9 205,2 167,6 177,7 157,5 167,6 [191,28] [197,28] [196,21] [179,00] [135,89] [144,09] [127,74] [135,25] # Survived periods 21,96 19,76 24,37 20,64 21,31 20,91 19,13 23,18 [7,12] [8,04] [5,95] [6,75] [7,08] [7,96] [7,50] [5,43]
# Obs. 1002 275 418 309 945 281 278 386
Std errors in brackets
TABLE 3: REGRESSION ANALYSIS
SIGMA=0.50 SIGMA=0.28 LD‐EAOC LD‐EPOC LD‐EAOC LD‐EPOCDumbreak ‐0.553*** ‐0.079 ‐0.538*** ‐0.139* [0.117] [0.071] [0.129] [0.067] Seq 1 0.057 0.050 0.150 0.048 [0.193] [0.117] [0.212] [0.109] Seq 2 ‐0.162 0.066 ‐0.075 ‐0.032 [0.181] [0.109 [0.198] [0.102] Period 0.445*** 0.195*** 0.545*** 0.243*** [0.037] [0.022] [0.041] [0.021] Period Squared ‐0.015*** ‐0.008*** ‐0.018*** ‐0.011*** [0.001] [0.001] [0.001] [0.001] Risk ‐0.061 ‐0.089*** ‐0.126** ‐0.055* [0.054] [0.032] [0.059] [0.031] Raven Test 0.201*** 0.009 0.182*** 0.031* [0.032] [0.019] [0.035] [0.018] Sex 0.664*** 0.179* 0.743*** 0.223** [0.168] [0.101] [0.184] [0.099] Constant 5.784*** 3.203*** 5.072*** 4.171*** [1.267] [0.766] [1.391] [0.720] Observations 1361 1361 1361 1361 R‐squared 0.18 0.09 0.16 0.15 Standard errors in brackets * significant at 10%; ** significant at 5%; *** significant at 1%
TABLE 4: BEHAVIORAL ANALYSIS SIGMA =0.50 SIGMA= 0.28 CONSUMPTION EXANTE EXPOST EXANTE EXPOSTOptimal consumption 0.101 0.888*** 0.228*** 0.524*** 0.006 0.453*** 0.131*** 0.348***
[0.129] [0.130] [0.079] [0.083] [0.063] [0.063] [0.046] [0.046]
Income 0.356*** 0.348*** 0.363*** 0.342***
[0.024] [0.023] [0.025] [0.023]
Available cash ‐0.009 0.002 ‐0.017* ‐0.001
[0.007] [0.007] [0.009] [0.007]
Observations 1361 1361 1361 1361 1361 1361 1361 1361
R‐squared 0.39 0.29 0.39 0.29 0.39 0.29 0.39 0.29
Standard errors in brackets * significant at 10%; ** significant at 5%; *** significant at 1%
24
FIGURE 1: CONSUMPTION IN TREATMENT LS10
050
100
150
Ave
rage
Con
s.
0 5 10 15 20 25 30period
Actual Cons. Ex-ante Cons. sig=0.50Ex-post Cons. sig= 0.50
Average Consumption (R=10)
050
100
150
Ave
rage
Con
s.
0 5 10 15 20 25 30period
Actual Cons. Ex-ante Cons. sig=0.28Ex-post Cons. sig= 0.28
Average Consumption (R=10)
25
FIGURE 2: CUMULATIVE SAVINGS IN TREATMENT LS10
050
100
150
Av.
Cum
. Sav
ings
.
0 5 10 15 20 25 30period
Actual Cum. Savs. Ex-ante Cum. Savs. sig=0.50Ex-post Cum. Savs. sig= 0.50
Av. Cum. Savings (R=10)
050
100
150
Av.
Cum
. Sav
ings
.
0 5 10 15 20 25 30period
Actual Cum. Savs. Ex-ante Cum. Savs. sig=0.28Ex-post Cum. Savs. sig= 0.28
Av. Cum. Savings (R=10)
26
FIGURE 3: CONSUMPTION IN TREATMENT LS15
050
100
150
Ave
rage
Con
s.
0 5 10 15 20 25 30period
Actual Cons. Ex-ante Cons. sig=0.50Ex-post Cons. sig= 0.50
Average Consumption (R=15)
050
100
150
Ave
rage
Con
s.
0 5 10 15 20 25 30period
Actual Cons. Ex-ante Cons. sig=0.28Ex-post Cons. sig= 0.28
Average Consumption (R=15)
27
FIGURE 4: CUMULATIVE SAVINGS IN TREATMENT LS15
050
100
150
Av.
Cum
. Sav
ings
.
0 5 10 15 20 25 30period
Actual Cum. Savs. Ex-ante Cum. Savs. sig=0.50Ex-post Cum. Savs. sig= 0.50
Av. Cum. Savings (R=15)
050
1001
50Av.
Cum
. Sav
ings
.
0 5 10 15 20 25 30period
Actual Cum. Savs. Ex-ante Cum. Savs. sig=0.28Ex-post Cum. Savs. sig= 0.28
Av. Cum. Savings (R=15)
28
Footnotes
1 See Gruber and Wise (1997) or Blondal and Scarpetta (1998). 2 Many studies have analyzed the relationship between retirement and Social Security. Earlier literature mainly focussed on the effect of the introduction of a pension system on the individual retirement decision, see among others, Feldstein (1977), Sheshinski (1978), Kotlikoff (1979a; 1979b), Crawford and Lilien (1981) or Cremer and Pestieau (2000). There is however more recent literature dealing with the retirement decision in a political economy environment, see Crettez and Le Maitre (2002), Conde-Ruiz et al. (2003; 2005), Conde-Ruiz and Galasso (2003; 2004), Casamatta et al. (2005), or Lacomba and Lagos (2006; 2007). 3 The link between lifetime contributions and benefits is being reinforced in a number of countries, Germany, Italy, Hungary, Mexico, Poland, Sweden, etc., by shifting from defined-benefit to defined-contribution systems. See Blondal and Scarpetta (1998). 4 An alternative way of raising the effective retirement age might be to delay the legal retirement age. If so, Lacomba and Lagos (2007) find that it would be appropriate to combine that measure with an increase in the redistributive character of the pension system. This higher intra-generational redistribution would delay the preferred legal retirement age of most of voters, which would result in a larger degree of approval in the postponement of the legal retirement age. 5 In a theoretical background it also requires that F(T)=0. 6 For the sake of simplicity we choose a zero discount rate. 7 Conditions for risk aversion are u′(c)>0 and u′′(c)<0. The Inada conditions imply that u′(0)=∞ and u′(∞)=0. The specific utility function for implementing the experiment is u(c) = 10 * Square Root (c). 8 For the sake of simplicity we choose a zero discount rate. 9 Reforms aiming to achieve actuarially fair social security systems must adjust pension benefits to achieve that the increase in pension benefits be exactly offset by the higher cost in terms of contributions and foregone pensions. 10 The electronic recruitment system at LINEEX follows a quite natural mechanism. The University of Valencia students get an email from the laboratory, announcing new sessions. They then go to a web based system to register in a given session, depending on their prior experience and the session requirements. The risk aversion test was the one by Holt and Laury (2002), based on a menu of ten paired lottery choices. Decisions in our experiment were matched with the decisions in the test by a common random code subjects used in both sessions. Identities were never linked to choices. The basic design is explained in the appendix. 11 As the conversion involves no uncertainty at all, this parameter is only a design artifact to generate a concave payoff function and allow for some consumption smoothing. Similar design features have been used in the related literature, described in section 2. 12 We denote by earliest periods the pre lump-sum rounds, that is, the 10 first rounds in LS10 and the 15 first rounds in LS15. 13 Additional socioeconomic variables, as Edumother and Edufather (controlling for the education level of mother and father), are excluded from the analysis as they proved to be never significant in all estimations. Full estimations are available from authors upon request. 14 Fixed effects are included to adjust for the possibility that some subjects saved more than others. The estimates of those variables are omitted for the sake of simplicity. Although not in the table, we have also performed the analysis by sequence for any of the treatments. Since the results do not change we only present results by pooled data.